Intelligent regulation and control method for AR content pushing

Abstract

The invention discloses an intelligent regulation and control method for AR content pushing, and the method comprises the following steps: S1, receiving AR content information according to the history of a pushing terminal, and selecting the content with the highest matching degree as the selected pushing content; if the user uses the push content for the first time, randomly selecting the push content; s2, collecting and analyzing hardware information of the terminal, and generating terminal hardware evaluation information; s3, carrying out resolution and layout adaptation on the selected push content according to the terminal hardware evaluation information, and generating preliminary adjustment push content; s4, collecting current real-time state information of the terminal, wherein the current real-time state information comprises a network bandwidth, a memory occupancy rate and a processor occupancy rate; and S5, based on the content complexity parameter and the real-time state information, generating final push content through a resource-content matching decision model. According to the method, through mechanisms such as hardware evaluation, resolution and layout adaptation, content complexity quantification, resource matching decision, real-time excitation optimization and the like, full-process dynamic regulation and control of AR content pushing are realized.

Description

Technical Field

[0001] This invention relates to the field of control methods, and more specifically to an intelligent control method for AR content push. Background Technology

[0002] AR content is interactive digital content that blends virtual information with real-world scenes. It achieves virtual-real combination display and interaction through hardware such as cameras and sensors on terminal devices. Its core characteristics include: high visual complexity: often includes 3D models, high-definition textures, dynamic UI elements, etc., requiring high terminal rendering capabilities; real-time interactivity: needs to update content in real time according to user actions and environmental changes, relying on the real-time response of terminal hardware and resources.

[0003] When AR content is pushed out, it faces challenges such as the diversity and performance differences of terminal hardware, dynamic changes in real-time resource status, and an imbalance between content complexity and resource matching.

[0004] Therefore, existing AR content push control processes suffer from problems such as lack of hardware-aware dynamic adaptation, absence of real-time resource optimization mechanisms, and disconnect between user experience and resource allocation. Thus, an intelligent control method for AR content push is proposed. Summary of the Invention

[0005] To address the shortcomings of existing technologies, this invention provides an intelligent control method for AR content push, comprising the following steps:

[0006] S1: Based on the AR content information received by the push terminal in the past, select the content with the highest matching degree as the selected push content; if the user is using it for the first time, the push content is randomly selected.

[0007] S2: Collect and analyze the hardware information of the push terminal to generate terminal hardware evaluation information;

[0008] S3: Based on the terminal hardware evaluation information, the selected push content is adapted for resolution and layout to generate initially adjusted push content;

[0009] S4: Collect the current real-time status information of the push terminal, including network bandwidth, memory usage and processor usage;

[0010] S5: Based on content complexity parameters and real-time status information, the final push content is generated through a resource-content matching decision model;

[0011] S6: When displaying the final push content, dynamically monitor memory and processor usage and perform real-time optimization through incentive analysis.

[0012] Furthermore, the process of obtaining the terminal hardware evaluation information is as follows:

[0013] Collect the GPU model, RAM size, screen resolution, and sensor type of the push terminal;

[0014] The GPU model is mapped to the rendering capability level G using a pre-defined hardware capability grading table. level RAM size is mapped to memory capacity level M level Screen resolution converted to pixel density D ppi and aspect ratio R aspect The sensor type is marked with the tracking accuracy indicator S. flag ;

[0015] Afterwards, regarding G level M level D ppi R aspect With S flag Integrate and generate terminal hardware evaluation information:

[0016] Terminal hardware evaluation information = {G level M level D ppi ,R aspect ,S flag}

[0017] Furthermore, the resolution and layout adaptation process includes resolution scaling, adaptive offset of layout anchor points, and rendering load control.

[0018] The specific process of dynamic resolution scaling is as follows: when D... ppi When <τ, the resolution adjustment mechanism is activated to adjust the resolution to the target resolution, where τ is a preset pixel density threshold based on the terminal hardware evaluation information;

[0019] The calculation process for the target resolution is as follows:

[0020] R target =η·D ppi ·R native ;

[0021] Where η is related to the rendering capability level G level Positively correlated scaling compensation coefficient;

[0022] The specific process of adaptive offset of layout anchor points is as follows:

[0023] According to the screen aspect ratio R aspect Calculate the offset of non-core UI elements based on the difference from the standard Rstd ratio, and adjust the position of non-core UI elements accordingly;

[0024] Calculate the offset of non-core UI elements, including horizontal and vertical offsets;

[0025] Horizontal offset: Δx =k x ·|R aspect -R std |;

[0026] Vertical offset: Δ y =k y ·|R aspect -R std |;

[0027] Where kx and ky are offset coefficients used to control the sensitivity of the offset;

[0028] The specific process of rendering load control is as follows:

[0029] When the rendering capability level Glevel is lower than the set threshold, a load optimization operation is performed;

[0030] The specific operations are as follows: merge layers to reduce the number of rendering layers and reduce GPU load;

[0031] Vector graphics bitmap conversion: Converting vector graphics into bitmaps reduces the amount of computation required for real-time rendering.

[0032] The scaling compensation coefficient η is determined according to the following rules:

[0033]

[0034] Among them, η1>η2>η3, and L1 and L2 are the rendering capability grading thresholds.

[0035] Furthermore, the generation of the content complexity parameter includes the following process:

[0036] Basic data extraction: Extract three core metrics for AR content, namely, the number of polygon faces P, texture resolution T, and the number of dynamic UI elements Nui;

[0037] Layout complexity factor calculation: The layout complexity factor U is calculated using the following formula:

[0038]

[0039] Complexity parameter quantification model: Calculate the content complexity parameter C by combining the number of polygon faces P, texture resolution T, and layout complexity factor U.

[0040]

[0041] Where α, β, γ are related to G level Negative correlation weighting coefficients, where Tmax is the preset maximum texture resolution;

[0042] The weight coefficients α, β, and γ are dynamically updated, and the specific dynamic update process includes:

[0043] Monitor the actual rendering latency of the final pushed content (Trender) and compare it with the preset threshold (Tmax);

[0044] If Trender > Tmax, increase the weight coefficient of the corresponding element according to the timeout ratio:

[0045] If texture rendering causes latency, increase β;

[0046] If polygon calculation causes a delay, increase α;

[0047] If layout complexity causes latency, increase γ.

[0048] Furthermore, the specific process of the resource-content matching decision model includes:

[0049] Calculate the real-time resource capacity Rc:

[0050]

[0051] μ is the memory weighting coefficient, M usage For memory usage, M level This refers to the memory capacity level (obtained through terminal hardware evaluation);

[0052] ν is the processor weight coefficient, P usage For processor utilization, C core This is a conversion factor for the number of processor cores;

[0053] κ is the network weight coefficient, B net Given the current network bandwidth, B max This is the theoretical maximum bandwidth;

[0054] Perform hierarchical matching decisions:

[0055] If the content complexity parameter C≤Rc, directly output the initial adjusted push content;

[0056] If C>Rc, initiate the degradation operation in priority order until C≤Rc is satisfied;

[0057] The resource capacity weight coefficients μ, ν, and κ are obtained through offline training:

[0058] Run AR content samples on multiple terminal devices and record the actual frame rate.

[0059] Using |Factual-Ftarget| as the loss function, μ, ν, and κ are optimized through gradient descent.

[0060] Furthermore, the priority order of the downgrade operations is as follows:

[0061] Level 1: Reduce texture resolution T and update texture-related terms in the content complexity parameter C;

[0062] Level 2: Reduce the number of dynamic UI elements (Nui) and update the layout complexity factor (U) related terms in the content complexity parameter (C);

[0063] Level 3: Perform layer merging and vector graphics bitmap conversion operations;

[0064] Level 4: Scale the target resolution Rtarget to σ times its original value (σ<1).

[0065] Furthermore, the process of the incentive analysis includes:

[0066] Define the fluency excitation function Q, and the specific process is as follows:

[0067]

[0068] Where k1 is the memory weight coefficient, M usage Mt represents memory usage, and Mt represents the memory safety threshold.

[0069] k2 is the processor weight coefficient, P usage Pt represents the processor utilization rate, and Pt represents the processor safety threshold.

[0070] Mt and Pt are determined by the memory capacity level M in the terminal hardware evaluation information. level The number of processor cores is dynamically calculated to reflect the carrying capacity of the terminal hardware.

[0071] Then, real-time optimization is performed based on the activation function, and the specific optimization content is as follows:

[0072] When the smoothness excitation function Q < the preset minimum value Qmin, the non-core layer shutdown mechanism is activated. Based on the user's gaze hotspot priority, the non-core layers that are farthest from the gaze area are shut down first to reduce system load.

[0073] Furthermore, the priority of user-focused hotspots is determined in the following way:

[0074] Real-time tracking of the coordinates of the area the user is looking at;

[0075] Calculate the distance Di between the center point of each UI layer and the viewing area;

[0076] Sort by Di in ascending order, and turn off the layer furthest away first.

[0077] Furthermore, the specific process of acquiring the real-time status information includes:

[0078] Before pushing the message, start the lightweight simulator, preload AR content, and record the peak memory and processor usage. Use the peak usage as supplementary input for real-time status information.

[0079] The beneficial effects of this invention are reflected in:

[0080] This intelligent control method for AR content push matches the push content based on the terminal's historical received information, and randomly selects the content on the first use to improve content relevance.

[0081] By collecting terminal hardware information to generate evaluation data, this system enables content resolution, layout adaptation, and rendering load control based on hardware capabilities. This ensures compatibility and performance optimization across different hardware devices. Resolution and UI layout are dynamically adjusted based on pixel density and screen aspect ratio to avoid display misalignment or blurring caused by screen parameter differences, thus improving the visual experience. When hardware rendering capabilities are insufficient, operations such as layer merging and vector bitmap conversion reduce GPU load, preventing stuttering due to excessive rendering pressure. Content complexity is quantified by extracting metrics such as polygon count, texture resolution, and dynamic UI elements. Weighting coefficients are dynamically allocated based on hardware capabilities, and rendering latency is monitored in real-time with adaptive weight adjustments. Targeted optimization of texture, polygon calculation, or layout complexity improves rendering efficiency. The source capacity model integrates real-time status data such as memory, processor, and network bandwidth to calculate the matching degree between content complexity and resources. When resources are insufficient, it performs degradation operations according to priority to ensure that AR content can still run smoothly in resource-constrained scenarios. It dynamically monitors memory and processor usage through a smoothness incentive function. When performance is insufficient, it closes non-core layers based on the user's gaze hotspot priority, focuses on the user's attention area, and improves interaction smoothness. Before pushing, it preloads AR content and collects peak resource usage to optimize resource allocation in advance and enhance system stability. The hardware safety threshold is dynamically calculated based on hardware evaluation information to adapt to the carrying capacity of different devices. The resource capacity weight coefficient is optimized through offline training, and parameters such as the scaling coefficient in degradation operations are preset to improve the model's adaptability to different scenarios. Attached Figure Description

[0082] To more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the accompanying drawings used in the description of the specific embodiments or the prior art will be briefly introduced below. In all the drawings, similar elements or parts are generally identified by similar reference numerals. In the drawings, the elements or parts are not necessarily drawn to scale.

[0083] Figure 1 This is the overall flowchart of the present invention. Detailed Implementation

[0084] The embodiments of the technical solution of the present invention will now be described in detail with reference to the accompanying drawings. These embodiments are merely illustrative of the technical solution of the present invention and are therefore intended to limit the scope of protection of the present invention.

[0085] It should be noted that, unless otherwise stated, the technical or scientific terms used in this application should have the ordinary meaning as understood by one of ordinary skill in the art to which this invention pertains.

[0086] like Figure 1 As shown, an intelligent control method for AR content push includes the following steps:

[0087] S1: Based on the AR content information received by the push terminal in the past, select the content with the highest matching degree as the selected push content; if the user is using it for the first time, the push content is randomly selected.

[0088] S2: Collect and analyze the hardware information of the push terminal to generate terminal hardware evaluation information;

[0089] S3: Based on the terminal hardware evaluation information, the selected push content is adapted for resolution and layout to generate initially adjusted push content;

[0090] S4: Collect the current real-time status information of the push terminal, including network bandwidth, memory usage and processor usage;

[0091] S5: Based on content complexity parameters and real-time status information, the final push content is generated through a resource-content matching decision model;

[0092] S6: When displaying the final push content, dynamically monitor memory and processor usage and perform real-time optimization through incentive analysis.

[0093] The process of obtaining the terminal hardware evaluation information is as follows:

[0094] Collect the GPU model, RAM size, screen resolution, and sensor type of the push terminal;

[0095] The GPU model is mapped to the rendering capability level G using a pre-defined hardware capability grading table. level RAM size is mapped to memory capacity level M level Screen resolution converted to pixel density D ppi and aspect ratio R aspect The sensor type is marked with the tracking accuracy indicator S. flag ;

[0096] Afterwards, regarding G level M level D ppi R aspect With Sflag Integrate and generate terminal hardware evaluation information:

[0097] Terminal hardware evaluation information = {G level M level D ppi R aspect S flag};

[0098] By collecting terminal hardware information and mapping it to rendering capability level, memory capacity level, pixel density, aspect ratio, and tracking accuracy, terminal hardware evaluation information is generated. This provides a data foundation for subsequent AR content adaptation, transforming hardware parameters into standardized levels and parameters. This allows the system to dynamically adjust the resolution and rendering load of AR content based on quantified hardware capabilities, avoiding display anomalies or performance lag caused by hardware differences. It covers the hardware characteristics of different terminal models, ensuring that AR content runs optimally on various devices through a unified evaluation model, expanding applicable scenarios. Based on memory capacity and rendering capability levels, it predicts device capacity in advance, avoiding resource waste, such as pushing highly complex content to low-configuration devices leading to crashes or performance redundancy, or underutilizing hardware performance on high-configuration devices.

[0099] For example, a user receives the same AR advertising content using two different terminal devices (phone A and AR glasses B).

[0100] Hardware Information Collection and Evaluation: Phone A: GPU model is Adreno 610, mapped to G level =2, RAM=4GB, M level =2, screen resolution 1080×2340, D ppi =409, R aspect =19.5:9, the sensor is a common gyroscope S flag = Low.

[0101] AR Glasses B: GPU model is Mali-G78, G level =4, RAM=8GB, M level =4, screen resolution 3840×2160, D ppi =1200, R aspect =16:9, sensor is a six-axis IMU + vision positioning, S flag =High.

[0102] Adaptation effect: For mobile phone A, with low hardware evaluation value, the system automatically reduces the texture resolution of AR content and merges rendering layers to avoid lag caused by excessive GPU load;

[0103] For AR glasses B, with high hardware evaluation values, the system upgrades the content resolution to 4K and enables more complex dynamic UI elements to fully utilize device performance and improve visual clarity and interaction smoothness.

[0104] The process of resolution and layout adaptation includes resolution scaling, adaptive offset of layout anchor points, and rendering load control.

[0105] The specific process of dynamic resolution scaling is as follows: when D... ppi When <τ, the resolution adjustment mechanism is activated to adjust the resolution to the target resolution, where τ is a preset pixel density threshold based on the terminal hardware evaluation information;

[0106] The calculation process for the target resolution is as follows:

[0107] R target =η·D ppi ·R native ;

[0108] Where η is related to the rendering capability level G level Positively correlated scaling compensation coefficient;

[0109] The specific process of adaptive offset of layout anchor points is as follows:

[0110] According to the screen aspect ratio R aspec t and standard ratio R std Calculate the offset of non-core UI elements based on the differences, and adjust the position of non-core UI elements accordingly;

[0111] Calculate the offset of non-core UI elements, including horizontal and vertical offsets;

[0112] Horizontal offset: Δ x =k x ·|R aspect -R std |;

[0113] Vertical offset: Δ y =k y ·|R aspect –R std |;

[0114] Where kx and ky are offset coefficients used to control the sensitivity of the offset;

[0115] The specific process of rendering load control is as follows:

[0116] When rendering capability level G level When the load falls below a set threshold, perform load optimization.

[0117] The specific operations are as follows: merge layers to reduce the number of rendering layers and reduce GPU load;

[0118] Vector graphics bitmap conversion: Converting vector graphics into bitmaps reduces the amount of computation required for real-time rendering.

[0119] The scaling compensation coefficient η is determined according to the following rules:

[0120]

[0121] Among them, η1>η2>η3, and L1 and L2 are rendering capability grading thresholds;

[0122] The resolution of AR content is dynamically adjusted based on the pixel density of the terminal to avoid blurry content on low-pixel-density devices or waste of resources on high-pixel-density devices, thus ensuring display accuracy on different screens.

[0123] Adjusting the position of non-core UI elements based on screen aspect ratio differences prevents UI elements from being obscured or incompletely displayed due to different screen ratios, thereby improving the smoothness of user operation.

[0124] When the terminal's rendering capabilities are insufficient, operations such as layer merging and vector bitmap conversion can be used to reduce the GPU load, prevent AR content from stuttering or crashing, and adapt to terminals with different hardware performance.

[0125] If a user uses mobile phone A (low hardware configuration) and AR glasses B (high hardware configuration) to browse the same AR shopping page, the page contains 3D product models, dynamic price tags, and navigation buttons.

[0126] Applications of dynamic resolution scaling:

[0127] Mobile Phone A: Dppi = 409 < preset threshold τ (e.g., 500), the system activates the resolution adjustment mechanism, according to formula R. target =η·D ppi ·R native Calculate the target resolution. Because of the G resolution of phone A... level =2, rendering capability is low. The scaling compensation factor η is set to η2 to reduce the original 4K resolution content to 1080P to avoid rendering delay due to excessively high resolution.

[0128] AR Glasses B:D ppi =1200≥τ, no need to reduce resolution, and because G level =4, η takes the value of η1, which can support higher resolution rendering and make the details of 3D product models clearer.

[0129] Application of adaptive offset for layout anchor points:

[0130] The aspect ratio R of phone A's screen aspect =19.5:9, standard ratio R std=16:9, the system calculates the offset of non-core UI elements (such as sidebar recommendation tags):

[0131] The horizontal offset Δx = kx * |19.5 / 9 - 16 / 9| = kx * 0.39, and the vertical offset Δy = ky * 0.39, shifts the label towards the center of the screen to avoid incomplete label display due to the screen being too narrow.

[0132] AR glasses B's R aspect =16:9 and R std Consistent, no offset required, UI elements are displayed according to the standard layout.

[0133] Furthermore, the generation of the content complexity parameter includes the following process:

[0134] Basic data extraction: Extract three core metrics for AR content, namely, the number of polygon faces P, texture resolution T, and the number of dynamic UI elements Nui;

[0135] Layout complexity factor calculation: The layout complexity factor U is calculated using the following formula:

[0136]

[0137] Complexity parameter quantification model: Calculate the content complexity parameter C by combining the number of polygon faces P, texture resolution T, and layout complexity factor U.

[0138]

[0139] Where α, β, γ are related to G level The weighting coefficient of negative correlation, T max This is the preset maximum texture resolution.

[0140] The weight coefficients α, β, and γ are dynamically updated, and the specific dynamic update process includes:

[0141] Monitor the actual rendering latency of the final pushed content (Trender) and compare it with the preset threshold (Tmax);

[0142] If Trender > Tmax, increase the weight coefficient of the corresponding element according to the timeout ratio:

[0143] If texture rendering causes latency, increase β;

[0144] If polygon calculation causes a delay, increase α;

[0145] If layout complexity causes latency, increase γ;

[0146] By extracting metrics such as polygon count, texture resolution, and number of dynamic UI elements from AR content, the system calculates layout complexity factors and content complexity parameters. Based on rendering latency, it dynamically adjusts weight coefficients to quantify and adaptively optimize content complexity. This transforms the visual and layout complexity of AR content into calculable parameters, enabling the system to predict content rendering pressure based on terminal hardware assessment information. This avoids device lag caused by overly complex content. Real-time monitoring of rendering latency and targeted weight adjustments allow the system to automatically focus on current performance bottlenecks, improving optimization efficiency. Regardless of whether AR content is 3D model-intensive, high-definition texture-intensive, or interactively complex, complexity can be quantified using a unified model, achieving universal adaptation.

[0147] For example, if the user uses mobile phone A, G level =2, low rendering capabilities and AR glasses B, G level =4, high rendering capability; when browsing the same AR museum exhibit page, the page includes:

[0148] 3D sculpture model, polygon count P = 100,000, texture resolution T = 4K;

[0149] Dynamic information labels, Nui=5, layer depth 3, screen coverage 20%.

[0150] Content complexity parameter calculation and optimization:

[0151] Basic data extraction: Mobile phone A and AR glasses B process the same content, P = 100,000, T = 4K, Nui = 5.

[0152] Content complexity parameter calculation and optimization:

[0153] Basic data extraction: Mobile phone A and AR glasses B process the same content, P = 100,000, T = 4K, Nui = 5.

[0154] Layout complexity factor U calculation: Screen area coverage = 20%, layer depth = 3, U = 5 × 3 / 20% = 75.

[0155] Complexity parameter C calculation: initial weight coefficients α, β, γ and G level Negative correlation, due to the G of mobile phone A level =2 (G lower than AR glasses B) level =4), its α, β, and γ values ​​are larger, meaning it is more sensitive to complexity;

[0156] Complexity parameter C calculation: initial weight coefficients α, β, γ and G level Negative correlation, due to the G of mobile phone A level＝ 2. G is lower than AR glasses B level =4, its α, β, γ values ​​are larger, that is, it is more sensitive to complexity;

[0157] Phone A:

[0158] AR Glasses B:

[0159]

[0160] Because of α A β A γ A Larger, the C calculated by phone A A C is higher than AR glasses B B The system predicted that mobile phone A would face greater rendering pressure.

[0161] Dynamic weight update (taking phone A as an example): After the push is sent, the actual rendering latency T of phone A is monitored. render =80ms>Preset threshold T max =60ms, determined to be latency caused by texture rendering; 4K textures are sensitive to low-G resolution. level When the equipment pressure is high, increase β according to the overtime ratio. A Weights.

[0162] After the weight update, the system prioritizes reducing the texture resolution, decreasing T from 4K to 2K, and then recalculates C. A At this point, the βA*(2K / 4K) term decreases, and C A Consequently, the resource capacity for adapting to mobile phone A decreases.

[0163] Optimization results:

[0164] Phone A: After reducing the texture resolution, the rendering latency dropped to 50ms, improving smoothness. Although the clarity of the 3D sculpture model decreased, there was no lag in the interaction.

[0165] AR Glasses B: Due to G level High, initial C B With a resource capacity of Rc, no downgrading is required; 4K textures and complex models are rendered normally with clear details.

[0166] For low-configuration devices, phone A: proactively identify rendering risks of highly complex content, dynamically adjust weights, and perform targeted downgrading, such as reducing textures, to ensure smoothness;

[0167] For high-configuration devices, AR glasses B: maintain the original complexity of the content, give full play to hardware performance, achieve on-demand adaptation, and improve the stability of AR experience and resource utilization on different devices.

[0168] The specific process of the resource-content matching decision model includes:

[0169] Calculate the real-time resource capacity Rc:

[0170]

[0171] μ is the memory weighting coefficient, M usage For memory usage, M level This refers to the memory capacity level (obtained through terminal hardware evaluation);

[0172] ν is the processor weight coefficient, P usage For processor utilization, C core This is a conversion factor for the number of processor cores;

[0173] κ is the network weight coefficient, B net Given the current network bandwidth, B max This is the theoretical maximum bandwidth;

[0174] Perform hierarchical matching decisions:

[0175] If the content complexity parameter C≤Rc, directly output the initial adjusted push content;

[0176] If C>Rc, initiate the degradation operation in priority order until C≤Rc is satisfied;

[0177] The resource capacity weight coefficients μ, ν, and κ are obtained through offline training:

[0178] Run AR content samples on multiple terminal devices and record the actual frame rate.

[0179] Using |Factual-Ftarget| as the loss function, μ, ν and κ are optimized by gradient descent, with Ftarget as a preset value;

[0180] Dynamic matching of resources across multiple dimensions improves system stability;

[0181] Resource capacity is calculated by combining real-time data such as memory usage, processor load, and network bandwidth to avoid misjudgments caused by single-dimensional evaluation (e.g., considering only memory while ignoring network latency), ensuring a balance between resource supply and demand during AR content runtime. Content complexity is gradually reduced according to preset priorities, prioritizing the preservation of core functions under resource constraints to avoid direct crashes or stuttering. Resource weight coefficients are optimized using frame rate loss functions across multiple devices, enabling the model to adapt to different hardware combinations and improving cross-device adaptation efficiency.

[0182] For example, if the user uses phone A, it has 4GB of RAM and a Snapdragon 660 processor;

[0183] AR glasses B, 8GB RAM, Snapdragon 888 processor;

[0184] Load the same AR game scene, the scene includes:

[0185] High-poly game character model, P=200,000, 4K texture, T=4K, dynamic skill effects, Nui=8.

[0186] Resource capacity Rc calculation and matching decision:

[0187] Resource capacity calculation for phone A:

[0188] Memory usage M usage =70%, M level =2, 4GB is mapped to level 2;

[0189] Processor utilization P usage =85%, C core =4, the conversion factor for quad-core processors;

[0190] Network bandwidth B net =10Mbps, B max =100Mbps;

[0191] Resource weight coefficients μ = 0.4, ν = 0.3, κ = 0.3, optimized for low- to mid-range devices through offline training;

[0192]

[0193] Resource capacity calculation for AR glasses B:

[0194] M usage =40%, M leve l=4, 8GB is mapped to level 4;

[0195] P usage =30%, C core =8, the conversion factor for an eight-core processor;

[0196] B net =50Mbps, B max =100Mbps;

[0197] Weighting coefficients μ = 0.3, ν = 0.5, κ = 0.2 (optimized for high-end equipment);

[0198]

[0199] Content complexity parameter C calculation:

[0200] Mobile phone A and AR glasses B process the same content. According to claim 4, C = 2.5 (hypothetical value, which needs to be greater than Rc = 0.45 for mobile phone A and less than Rc = 3.62 for AR glasses B).

[0201] Tiered downgrade operation (taking phone A as an example):

[0202] Since C = 2.5 > Rc = 0.45, initiate degradation according to priority: First level: Reduce the texture resolution, lower 4K texture to 1K, update the texture entry in C, and recalculate C = 2.5 - β·(4K - 1K) / 4K ≈ 2.0 (assuming β = 0.5);

[0203] Second level: Reduce the number of dynamic UI elements, decrease Nui from 8 to 4, update the layout complexity factor U, and calculate C = 2.0 - γ·(8 - 4)·layer depth / coverage ≈ 1.5;

[0204] Third level: Merge layers and rasterize vector graphics, merge the lighting layers of the game character model, reduce the number of rendering layers, and further reduce C to 1.0;

[0205] At this time, C = 1.0 is still > Rc = 0.45, initiate the fourth level: Reduce the resolution, scale the target resolution to σ = 0.5 times the original value, and finally reduce C to 0.8 ≤ Rc = 0.45 (assuming that σ meets the conditions after adjustment).

[0206] Mobile phone A: Through four levels of degradation, although the texture clarity and UI animation effects are reduced, the core display of the game character model is normal, and the game frame rate is increased from 15fps to 28fps, ensuring basic interaction fluency;

[0207] AR glasses B: Since C = 2.5 < Rc = 3.62, no degradation is required, 4K texture and all special effects are rendered normally, and the frame rate is stable at 60fps, providing a smooth experience.

[0208] The priority order of the degradation operations is as follows:

[0209] First level: Reduce the texture resolution T, and update the texture-related items in the content complexity parameter C;

[0210] Second level: Reduce the number of dynamic UI elements N ui , and update the layout complexity factor U-related items in the content complexity parameter C;

[0211] Third level: Perform layer merging operations and vector graphic rasterization operations;

[0212] Fourth level: Scale the target resolution R target to σ times the original value (σ < 1).

[0213] The process of the incentive analysis includes:

[0214] Define the fluency incentive function Q, and its specific process is as follows:

[0215]

[0216] Among them, k1 is the memory weight coefficient, M usageMt represents memory usage, and Mt represents the memory safety threshold.

[0217] k2 is the processor weight coefficient, P usage Pt represents the processor utilization rate, and Pt represents the processor safety threshold.

[0218] Mt and Pt are determined by the memory capacity level M in the terminal hardware evaluation information. level The number of processor cores is dynamically calculated to reflect the carrying capacity of the terminal hardware.

[0219] Then, real-time optimization is performed based on the activation function, and the specific optimization content is as follows:

[0220] When the smoothness excitation function Q < the preset minimum value Qmin, the non-core layer shutdown mechanism is activated. Based on the user's gaze hotspot priority, the non-core layer that is farthest from the gaze area is shut down first to reduce system load.

[0221] By defining a smoothness excitation function Q and combining memory usage, processor usage, and hardware security thresholds, the performance status of AR content during runtime is dynamically monitored. When the smoothness is insufficient, non-core layers are turned off based on the priority of user attention hotspots, achieving a dual optimization of focusing on the user's attention area and reducing system load.

[0222] The system dynamically calculates smoothness based on real-time memory and processor usage, proactively identifying performance bottlenecks. In cases of memory overflow or CPU overload, it quickly releases resources by disabling non-core layers, preventing AR content from stuttering or crashing.

[0223] The layer closing priority is determined based on the user's focus area, prioritizing the preservation of content in the user's attention area while sacrificing non-core elements in non-focus areas, such as background decorations and edge advertisements, thus ensuring the core interactive experience while optimizing performance.

[0224] The memory safety threshold Mt and processor safety threshold Pt are dynamically calculated based on terminal hardware evaluation information. For example, higher utilization rates are allowed for high-configuration devices to avoid one-size-fits-all optimization and adapt to the carrying capacity of different devices.

[0225] For example, user A has 4GB of memory, G... level =2, browse the AR e-commerce page, the page includes:

[0226] Central main product 3D model, core layer, user-focused area;

[0227] Side panel recommended product list (non-core layer), bottom navigation bar (semi-core layer), background dynamic effects (non-core layer).

[0228] Smoothness activation function calculation and optimization:

[0229] Dynamic calculation of hardware security threshold: Memory capacity level M of mobile phone A level = 2, the number of processor cores is 4 cores, calculated by a preset formula: Memory security threshold Mt = 70%, more memory space is reserved for low - configuration devices;

[0230] Processor security threshold P t = 80%, to avoid overheating caused by high load.

[0231] Calculation of smoothness incentive function Q: When browsing, it is monitored that M usage = 85% > Mt = 70%, P usage = 85% > P t = 80%, substitute into the formula:

[0232]

[0233] The preset minimum value Qmin = 0.7, because Q = 0.63 < Qmin, start the non - core layer closing mechanism.

[0234] Determination of user gaze hot - spot priority: Real - time tracking of the coordinates of the user's gaze area (central main product position), calculate the distance from the center point of each layer: Main product model: Di = 0 (gaze center);

[0235] Side - bar recommendation list: Di = 150 pixels;

[0236] Bottom navigation bar: Di = 80 pixels;

[0237] Background special effect: Di = 200 pixels;

[0238] Sorted in ascending order of Di, closing priority: Background special effect (Di is the largest) > Side - bar recommendation list > Bottom navigation bar, semi - core, not closed temporarily.

[0239] Layer closing and performance optimization: Close the background dynamic special effect and the side - bar recommendation list, release about 200MB of memory, and the processor load drops;

[0240] Recalculate Q = 0.63 + the gain brought by resource release (assumed to be increased to 0.72 ≥ Qmin), and smoothness is restored.

[0241] In terms of performance: The memory occupancy rate drops from 85% to 65%, the processor occupancy rate drops from 85% to 75%, and the AR page frame rate increases from 20fps to 30fps;

[0242] In terms of experience: The main product model gazed at by the user is always clearly displayed, only the edge non - core content is closed, and the core interactions are not affected, such as product rotation and detail viewing.

[0243] The user gaze hot - spot priority is determined by the following method:

[0244] Real-time tracking of the coordinates of the area the user is looking at;

[0245] Calculate the distance Di between the center point of each UI layer and the viewing area;

[0246] Sort by Di in ascending order, and turn off the layers furthest away first;

[0247] By tracking the coordinates of the user's gaze area in real time, calculating the distance Di between the center point of each UI layer and the gaze area, and sorting them in ascending order of Di, the non-core layers that are furthest away are turned off first, thus achieving dynamic resource optimization based on user attention.

[0248] Dynamically adjust the layer display priority based on the point of focus to ensure that the core content that users are interested in is always visible, such as the central product model and interactive buttons, thereby improving interaction efficiency and immersive experience.

[0249] Automatically identify layers in areas of non-user interest and prioritize turning them off, such as edge ads and background decorations, to reduce system load without affecting the core experience and adapt to scenarios with insufficient hardware performance.

[0250] Layer priority is updated in real time as the user's gaze moves, supporting interaction needs in multiple scenarios and improving the naturalness of AR interaction.

[0251] The specific process of collecting the real-time status information includes:

[0252] Before pushing the message, start the lightweight simulator, preload AR content and record the peak memory and processor usage, and use the peak usage as a supplementary input for real-time status information.

[0253] By preloading AR content and collecting peak resource usage, such as peak memory and peak processor load, the potential pressure of content on the terminal can be identified before push, avoiding lag or crashes caused by insufficient resources during real-time operation and improving system stability.

[0254] Real-time status information typically reflects the current operating load, while pre-loaded peak data can supplement resource requirements in extreme scenarios, enabling the system to more comprehensively assess the compatibility of content and devices and avoid making misjudgments based solely on the current status.

[0255] The lightweight simulator completes resource stress testing before pushing content, eliminating the need for high-load real-time evaluation during content runtime, thus reducing terminal computing overhead, making it especially suitable for devices with limited hardware performance.

[0256] Based on peak usage data, the system can dynamically adjust the content complexity at the beginning of the push, avoiding initial lag caused by pushing first and then optimizing, and improving the smoothness of the user experience.

[0257] The memory safety threshold Mt and processor safety threshold Pt are dynamically calculated based on the memory capacity level Mlevel and the number of processor cores in the terminal hardware evaluation information, reflecting the device's hardware capacity. The following example uses mobile phone A and AR glasses B:

[0258] Phone A (low configuration):

[0259] RAM = 4GB, mapped to memory capacity level M level =2, assuming 4GB corresponds to level 2 and 8GB corresponds to level 4;

[0260] Number of processor cores = 4 cores.

[0261] AR Glasses B, High Configuration:

[0262] RAM = 8GB, mapped to memory capacity level M level =4;

[0263] Number of processor cores = 8 cores.

[0264] Dynamic calculation method:

[0265] The memory safety threshold Mt is calculated as follows: Mt = Mtbase + (Mlevel × ΔMt);

[0266] Settings: Mtbase = 50% (basic security threshold), ΔMt = 10% (increment per level);

[0267] For mobile phone A, Mt = 50% + (2 × 10%) = 70%;

[0268] AR glasses B, Mt = 50% + (4 × 10%) = 90%.

[0269] Calculation of processor security threshold Pt:

[0270] Basic formula, P t =P tbase +(Number of cores × ΔP) t );

[0271] Set Ptbase = 60% (basic security threshold) and ΔPt = 3.75% (increment per core).

[0272] For mobile phone A, Pt = 60% + (4 × 3.75%) = 75%;

[0273] AR glasses B, Pt = 60% + (8 × 3.75%) = 90%.

[0274] Mt and M level Positive correlation: The higher the memory capacity level (e.g., the M of AR glasses B), the better. level=4), the higher the memory utilization rate that the device can handle, the higher Mt increases from 70% to 90%.

[0275] Pt is positively correlated with the number of processor cores: the more cores, such as the 8 cores in AR glasses B, the stronger the processor's parallel computing capability, and the higher the safety threshold Pt is from 75% to 90%.

[0276] Mobile phone A, Mt = 70%, Pt = 75%;

[0277] When memory usage M usage When Mt = 70%, the smoothness excitation function Q is calculated. If Q is lower than the threshold, the non-core layer is turned off.

[0278] AR glasses B, Mt=90%, Pt=90%;

[0279] Allow for higher resource utilization when M usage =85% is still within the safe range, no optimization is required, ensuring smooth operation of highly complex AR content.

[0280] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features therein. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention, and they should all be covered within the scope of the claims and specification of the present invention.

Claims

1. A smart control method for AR content push, characterized in that, Includes the following steps: S1: Based on the historical AR content information received by the push terminal, select the content with the highest matching degree as the selected push content; If the user is a first-time user, the push content will be randomly selected; S2: Collect and analyze the hardware information of the push terminal to generate terminal hardware evaluation information; S3: Based on the terminal hardware evaluation information, the selected push content is adapted for resolution and layout to generate initially adjusted push content; S4: Collect the current real-time status information of the push terminal, including network bandwidth, memory usage and processor usage; S5: Based on content complexity parameters and real-time status information, the final push content is generated through a resource-content matching decision model; S6: When displaying the final push content, dynamically monitor memory and processor usage and perform real-time optimization through incentive analysis.

2. The intelligent control method for AR content push according to claim 1, characterized in that: The process of obtaining the terminal hardware evaluation information is as follows: Collect the GPU model, RAM size, screen resolution, and sensor type of the push terminal; The GPU model is mapped to the rendering capability level G using a pre-defined hardware capability grading table. level RAM size is mapped to memory capacity level M level Screen resolution converted to pixel density D ppi and aspect ratio R aspect The sensor type is marked with the tracking accuracy indicator S. flag ; Afterwards, regarding G level M level D ppi R aspect With S flag Integrate and generate terminal hardware evaluation information: Terminal hardware evaluation information = {G level M level D ppi R aspect S flag } 3. The intelligent control method for AR content push according to claim 2, characterized in that: The process of resolution and layout adaptation includes resolution scaling, adaptive offset of layout anchor points, and rendering load control. The specific process of dynamic resolution scaling is as follows: when D... ppi When <τ, the resolution adjustment mechanism is activated to adjust the resolution to the target resolution, where τ is a preset pixel density threshold based on the terminal hardware evaluation information; The calculation process for the target resolution is as follows: R target =η·D ppi ·R native ; Where η is related to the rendering capability level G level Positively correlated scaling compensation coefficient; The specific process of adaptive offset of layout anchor points is as follows: According to the screen aspect ratio R aspect Calculate the offset of non-core UI elements based on the difference from the standard Rstd ratio, and adjust the position of non-core UI elements accordingly; Calculate the offset of non-core UI elements, including horizontal and vertical offsets; Horizontal offset: Δx = k x ·|R aspect -R std |; Vertical offset: Δy = k y ·|R aspect -R std |; Where kx and ky are offset coefficients used to control the sensitivity of the offset; The specific process of rendering load control is as follows: When rendering capability level G level When the load falls below a set threshold, perform load optimization. The specific operations are as follows: merge layers to reduce the number of rendering layers and reduce GPU load; Vector graphics bitmap conversion: Converting vector graphics into bitmaps reduces the amount of computation required for real-time rendering. The scaling compensation coefficient η is determined according to the following rules: Among them, η1>η2>η3, and L1 and L2 are the rendering capability grading thresholds.

4. The intelligent control method for AR content push according to claim 3, characterized in that: The generation of the content complexity parameter includes the following process: Basic data extraction: Extract three core metrics for AR content, namely, the number of polygon faces P, texture resolution T, and the number of dynamic UI elements Nui; Layout complexity factor calculation: The layout complexity factor U is calculated using the following formula: Complexity parameter quantification model: Calculate the content complexity parameter C by combining the number of polygon faces P, texture resolution T, and layout complexity factor U. Where α, β, γ are related to G level Negative correlation weighting coefficients, where Tmax is the preset maximum texture resolution; The weight coefficients α, β, and γ are dynamically updated, and the specific dynamic update process includes: Monitor the actual rendering latency of the final pushed content (Trender) and compare it with the preset threshold (Tmax); If Trender > Tmax, increase the weight coefficient of the corresponding element according to the timeout ratio.

5. The intelligent control method for AR content push according to claim 1, characterized in that: The specific process of the resource-content matching decision model includes: Calculate the real-time resource capacity Rc: μ is the memory weighting coefficient, M usage For memory usage, M level Memory capacity level; ν is the processor weight coefficient, P usage For processor utilization, C core This is a conversion factor for the number of processor cores; κ is the network weight coefficient, B net Given the current network bandwidth, B max This is the theoretical maximum bandwidth; Perform hierarchical matching decisions: If the content complexity parameter C≤Rc, directly output the initial adjusted push content; If C>Rc, initiate the degradation operation in priority order until C≤Rc is satisfied; The resource capacity weight coefficients μ, ν, and κ are obtained through offline training: Run AR content samples on multiple terminal devices and record the actual frame rate. Using |Factual-Ftarget| as the loss function, μ, ν, and κ are optimized through gradient descent.

6. The intelligent control method for AR content push according to claim 5, characterized in that: The priority order of the downgrade operations is as follows: Level 1: Reduce texture resolution T and update texture-related terms in the content complexity parameter C; Level 2: Reduce the number of dynamic UI elements (Nui) and update the layout complexity factor (U) related terms in the content complexity parameter (C); Level 3: Perform layer merging and vector graphics bitmap conversion operations; Level 4: Scale the target resolution Rtarget to σ times its original value.

7. The intelligent control method for AR content push according to claim 1, characterized in that: The process of the incentive analysis includes: Define the fluency excitation function Q, and the specific process is as follows: Where k1 is the memory weight coefficient, M usage Mt represents memory usage, and Mt represents the memory safety threshold. k2 is the processor weight coefficient, P usage Pt represents the processor utilization rate, and Pt represents the processor safety threshold. Mt and Pt are determined by the memory capacity level M in the terminal hardware evaluation information. level The number of processor cores is dynamically calculated to reflect the carrying capacity of the terminal hardware. Then, real-time optimization is performed based on the activation function, and the specific optimization content is as follows: When the smoothness excitation function Q < the preset minimum value Qmin, the non-core layer shutdown mechanism is activated. Based on the user's gaze hotspot priority, the non-core layers that are farthest from the gaze area are shut down first to reduce system load.

8. The intelligent control method for AR content push according to claim 7, characterized in that: The priority of user-focused hotspots is determined in the following way: Real-time tracking of the coordinates of the area the user is looking at; Calculate the distance Di between the center point of each UI layer and the viewing area; Sort by Di in ascending order, and turn off the layer furthest away first.

9. The intelligent control method for AR content push according to claim 1, characterized in that: The specific process of collecting the real-time status information includes: Before pushing the message, start the lightweight simulator, preload AR content, and record the peak memory and processor usage. Use the peak usage as supplementary input for real-time status information.

Images